Scripted replies for end-to-end tests

For Ghostwriter's own browser tests (the ghostwriter-e2e repository), the addon can answer every model call from a scripted scenario instead of a provider, so a test can drive the real control panel without spending tokens. It is for local development sites only and is off unless you set it up.

When it is on#

All of these must hold:

• GHOSTWRITER_FAKE_SCENARIOS in .env (config ghostwriter.testing.fake_scenarios) is set to the folder of scenario files. It is unset by default.
• APP_ENV is local or testing (never production).
• The request names a scenario: the X-Ghostwriter-Fake: <scenario> header, or the ghostwriter_fake cookie. The value is <name> or <name>#<run>, where the name is a file under the folder without .json (lower-case letters, digits, -, _ and / only).

Requests without the header or cookie are untouched, so an editor working on the same site in their own browser still gets the real provider. A name that doesn't match a file stops the request with an error rather than falling back to a real provider.

Every job queued during such a request carries the scenario in its payload (ghostwriterFake); the queue worker plays it for that job and goes back to the real providers afterwards.

Scenario files#

Scenarios are JSON, read by core's FakeScenario (see core's docs/providers.md): replies per agent (writer, brief-filler, layout-planner, reviewer, verifier and so on), handed out in order across requests and jobs and counted in the cache per run. Agents a scenario doesn't list are answered from their schema.

Setting it up on a test site#

GHOSTWRITER_FAKE_SCENARIOS=/Users/you/Dev/ghostwriter-e2e/scenarios

Restart the queue worker after changing it (php artisan queue:restart). Never set it on a staging or production site. The code is in src/Testing/FakeScenarios.php.